Healthcare AI is drowning in vendor promises and inflated ROI claims. Last week, Aboard, a premier provider of AI-accelerated software development, brought together professionals who’ve actually been in the room when AI projects succeed, and when they spectacularly fail.
Aboard hosted a roundtable that cut through the noise to answer the question keeping healthcare leaders up at night: when does AI actually move the needle on quality, cost, or access?
Moderated by Paul Ford (Co-founder at Aboard), the conversation featured Chethan Sarabu (Director of Clinical Innovation at Cornell Tech’s Health Tech Hub), Kamal Menghrajani (Former Assistant Director for Cancer Innovation & Public Health at The White House), and Erynn Petersen (CEO & Co-founder of Emme). The timing couldn’t have been better: OpenAI had just announced its push into healthcare AI, making the conversation even more urgent.
As Chethan noted, New York is becoming the epicenter for AI in healthcare. But the real question isn’t where the tech is happening—it’s where it should and shouldn’t be deployed. The event’s discussion was structured around the four critical ‘M’s shaping the healthcare innovation conversation: Metrics (and Measurement), Money, Mess (worst-case scenarios), and Merit (what’s working).
Metrics That Matter (Not the Ones in Press Releases)
The panel kicked off with the most fundamental question: how do you actually measure success in healthcare AI?
“It depends on what you’re going for,” Kamal explained. “What is the endpoint, the outcome? Then you design the study to get there.” She pointed to trial matching services as an example—AI-powered tools where the real outcome isn’t the technology itself, but the number of patients getting into appropriate clinical trials. The metric shapes the AI, not the other way around.
For Erynn at Emme, the metrics are refreshingly straightforward: How much money are we saving patients? How many are returning for care? Are people getting treatment they weren’t getting before? She shared a powerful example of someone who had cancelled an appointment because they thought they couldn’t afford care and were going to skip the doctor entirely. “Early on at Amazon, the question was: how do we make it as cheap as possible for people to get what they want,” she said. That same principle applies to healthcare.
But measuring AI in healthcare gets tricky fast. As Paul noted, “Not all that is measured matters, and not all that matters is measured.” The FDA hasn’t fully figured out how to regulate predictive models yet. OpenAI recently released a health benchmark that relies only on clinician evaluators—but what about patient evaluators? When someone asks, “why am I dizzy?” the way a patient frames that question is fundamentally different from how a clinician might ask it. LLMs are particularly profound in healthcare because so much of medicine happens in the realm of language. You might receive numerical blood test results, but how your brain processes that information is verbal. AI opens up new ways of measuring what matters—including things like the level of empathy from a physician.
And sometimes, Kamal reminded us, you just have to pivot. AI can’t replace community. When that becomes clear, the smart move is to change direction.
Money (And Who’s Willing to Wait for It)
The second M, money, brought the conversation to what hospitals and institutions actually care about: the bottom line.
Kamal shared an interesting development: “ChatGPT for healthcare—one of the things happening right now is just a one-year trial period. No long-term contract. One year to prove you can make our institution more money. That is what hospitals and institutions care about.” The question is whether ChatGPT has advanced enough to deliver on that promise. Right now, we’re in the test period.
The panel discussed the broader broken system and the radical transformation underway. Tools like OpenHand are helping with visit preparation and other administrative tasks. But Erynn was skeptical about some approaches: “Ten years ago, I was in meetings with big media companies who said ‘Facebook will give us their data.’ No, they were lying. Companies that believed that and didn’t follow the transformation are gone now.”
She drew a sharp distinction: “That’s not medical care, that’s a consumer service”—pointing to companies like Hims and Hers. The difference matters.
The conversation touched on how to realign incentives and concentrate quality in care delivery. As Chethan put it, New York is the best place for health tech based on both metrics and vibes—and combining that with new frameworks for examining how we deliver care could be transformative.
Mess (The Worst-Case Scenarios)
Paul asked the panel what keeps them up at night about AI in healthcare. The answers were sobering.
“The loss of expertise, the loss of faith in expertise,” one panelist said. “There’s an erosion happening. ChatGPT for health is a slippery slope—it’ll give you a quick answer but no context or nuance. I don’t think LLMs will replace us. These jobs aren’t going away. But I worry about public perception that we won’t need these doctors.”
Erynn raised two concerns: one about AI itself, and one about investment patterns affecting institutions and hospitals. “It’s not primary care physicians creating the vacuum,” she noted, pointing to broader systemic issues.
The panel discussed an underappreciated risk: removing friction from healthcare can actually be dangerous. Getting a refill used to require going to the doctor—and during that visit, something else might come up that you’d otherwise miss. That’s a core function of the human physician.
Paul pushed back on the idea that AI should make you smarter than your doctor: “Bad stuff can happen from that. The outcome will be better if you accept some friction.”
There’s also the challenge of data, both real-world data and synthetic data generation. Who owns it? How is it used? What are the guardrails?
As Paul noted, advocating for the relationship between patient and provider becomes a political conversation. But it’s one worth having.
Merit (Who’s Actually Doing This Well)
The final M focused on what’s working right now.
Kamal highlighted the use of AI agents, pointing to Grove AI’s work in clinical trials and biopharma to extend reach. “There’s friction—I don’t want to call my doctor—but if there’s an AI agent that helps with that, it could be valuable.” She emphasized that the government can only do so much to regulate. “We need to think about how we want healthcare AI to function and build systems that will improve our health. There aren’t a lot of guardrails right now.”
Erynn mentioned the importance of community, pointing to physician Slack groups and resources like Out of Pocket Health. She also highlighted Carter Dredge’s work on changing the economics of healthcare.
The panel discussed OpenNotes and other research groups working to make healthcare more transparent and accessible to patients.
The Bottom Line
The conversation reinforced what we already suspected: healthcare AI isn’t about replacing doctors or making flashy diagnoses. It’s about reducing administrative burden, leveraging data for patient benefit, and solving actual problems rather than creating new headaches.
The real opportunity lies in cutting through bureaucracy, not in replacing human judgment. And the companies that will succeed are the ones that understand the difference between a consumer service and actual medical care.
As the panel made clear, we’re still in the early stages. The test period is underway. And the organizations that approach AI with clear metrics, realistic timelines, and a commitment to protecting what matters—the patient-provider relationship—will be the ones that actually deliver on AI’s promise in healthcare.
– Sammy Jordan, VP